Sampling with Probabilities Proportional to Size: Adjustment for Changes in the Probabilities

Nathan Keyfitz

Journal of the American Statistical Association · 1951 · 57 citations · 0 references

Concepts

Abstract

W E CONSIDER a simplified area sample consisting of one primary sampling unit chosen from each of a number of strata; each such selected unit is to be enumerated completely. The sample (i.e., the total in the sample of whatever is being surveyed) can be converted to an estimate of the population in each stratum simply by multiplying by the number of units in the stratum; and this is, for practical purposes, the only method of estimating if no information about the unsampled units other than their number is on hand. If, however, some evidence is available on the relative sizes of the units in a stratum one would like to incorporate this in the estimating procedure. In an application which has now become fairly common, a number of different characteristics are to be estimated by a uniform procedure from a single survey, and the measure of size related to all characteristics is the number of persons in the several units at a precedingcensus. The Canadian Labor Force sample, like that used for the Monthly Report on the Labor Force of the United States, consists of up to four successive stages;' the theory here considered is applicable to the first stage. In the first stage the entire rural territory of Canada was divided into about 500 areas constituting primary sampling units, and these were assembled into strata each containing between 5 and 10 units. There was no requirement that the units comprising a stratum were to be contiguous; but only that they be as similar as possible in respect of type of farming, density of population etc.; nor was any attempt made to assemble into one stratum those units similar in the numbers of people they contained at the preceding census. The sub-sampling within selected primary sampling units was arranged so that multiplication of the sample take by 100 furnished an unbiased estimate of the stratum total; this was accomplished, following the suggestion of Hansen and Hurwitz, by sub-sampling in a ratio which depended on which unit was selected, so as to give a fixed expected number in the sample, i.e., 0.01 of the measure of size of the stratum. The application of the theorem of this paper, however, is not affected by sub-sampling; we may think of the estimating procedure as a matter of multiplying